KOS Design for Healthcare Decision-making

StrategiesTaxonomy
October 13, 2016 Copyright 2015 Taxonomy Strategies LLC. All rights reserved.
KOS Design for Healthcare Decision-making Based
on Consumer Criteria and User Stories
Joseph A. Busch, Principal Researcher, Taxonomy
Strategies, Washington, D.C.
Vivian Bliss, Independent Contractor, Kirkland, WA

2Taxonomy Strategies The business of organized information
Consumer Health Care Taxonomy background
 Designed to support types of queries a consumer health care information service
such as a website might get from a wide variety of consumers in a wide variety of
care conditions.
 Project sponsor:
 U.S. Centers for Medicare and Medicaid Services (CMS)
 Users:
 Medicare/Medicaid beneficiary*
 Caregiver
* Medicare is the U.S. government single payer health insurance for seniors over 65 years old. Medicaid is the
U.S. jointly funded federal and State health insurance program for low-income people.

Related research: Information seeking
 Critique of social science qualitative methods (Davenport). Scientificity – consumer decision
making is very different from studies of students, engineers and scientists.
 Most consumers search for health information on the Internet, usually starting with an organic
search engine. The most commonly researched topics are diseases or conditions, treatments
or procedures, and doctors or other health professionals. Half of online health information
research is on behalf of someone else. (Pew)

Related research: Quality of care
 Evidence-based decision-making by clinicians vs. factors that patients identify as most
important such as cost, qualifications and accessibility of care. (Hibbard & Sofaer)
 Patient narratives are of more interest to consumers, and easier for them to understand.
(Schlesinger)

Related research: KOS development
 Most health care KOS were originally designed to support researchers, clinicians and health
insurers. But they can be useful sources to build consumer-oriented health care KOS, rather
than starting from scratch. (Hyvönen)
 Consumer terminology used in health care related activities can be useful to improving
existing health care KOS. (Doing-Harris)

Sources: CMS health care website prototype
CareFinder prototype envisions leveraging Medicare’s massive datasets to support consumers in
making better health care decisions.

Sources: Medicare “Compare” websites and datasets
Compare websites and datasets provide directory information about CMS-registered service
providers and suppliers, and reported quality measures.

Sources: Inventory of data.medicare.gov datasets
Each Medicare dataset has a different structure and number of tables.

Sources: Inventory of quality measures
Existing quality measures were categorized by type, usefulness, availability and source.

Sources: Existing CMS and HHS taxonomies
The methods and learnings from earlier projects helped inform the Consumer Health Care
Taxonomy approach.
Substance Abuse and Mental Health Services
Administration (SAMHSA)
Healthcare.gov 

Sources: Authoritative sources, websites and query logs
 More than 100 vocabulary sources for Consumer Health Care concepts including:
 National Library of Medicine’s Medical Subject Headings (MeSH).
 International Classification of Diseases (ICD) used by CMS and other health insurance providers to
classify diseases and conditions.
 Unified Medical Language System (UMLS), a mapping of more than 100 vocabularies and
classification systems including MeSH and ICD.
 Query logs from Physician Compare and MedLine Plus to help identify unique facets, relevant
terms, and synonyms for the Taxonomy.
 Semantic relationships between Consumer Health Care Taxonomy concepts were made
based on trusted sources such as:
 Online symptom checkers from the Mayo Clinic, Cleveland Clinic, NHS UK, and HealthDirect
Australia.
 Physician Compare mappings of conditions and symptoms to medical specialties.
 Google medical search.

Sources: Interviews with SMEs and key stakeholders
 More than 30 CMS staff working on the various Compare websites.
 CMS Compare website data contractors.
 IDEO team that built the CareFinder prototype.
 CMS user research staff.
 External health care professionals.
 Friends and family who have health care stories.

What do consumers want or need to make choices
about where to get care?

Sources: User stories
Sharon is a 52 year old with ESRD who
received a kidney transplant 4 years ago. The
kidney has recently begun to fail and she
needs hemodialysis again 3 times a week. She
also recently broke her leg and needs physical
therapy 2 times a week. She works as a
waitress but is currently unemployed due to
her accident, and does not have insurance
besides Medicaid. Sharon needs to find a
dialysis center close to home since a family
member will need to drive her due to her leg
injury. She also needs to coordinate her
dialysis (3x week) with her physical therapy (2x
week).
Paula is an 85 year-old woman. She was out
driving in her neighborhood when all of a
sudden she lost her way and she couldn’t find
her way back home. This episode scared her
and her family. She began to worry about her
mental capacity, and wondered what kind of
specialist she could see who could assess her
mental acuity. Paula needs to figure out what
type of specialist can help assess her mental
acuity, and find a trustworthy specialist who is
close to her apartment and covered by her
health plan.

Consumer Health Care Taxonomy: Purpose
Needs to function as middleware that translates consumer queries into the language necessary
for retrieval of data from Medicare.gov datasets and Good to Know (GTK) content.
Compare
Data
Good to
Know
Consumer Health
Care Taxonomy
TechnicalLanguage
ConsumerQueries
Comprehensive &
accurate results
Highly relevant
& timely content
Search Engine
Results

Consumer Health Care Taxonomy: Functional requirements
 Provide enough information for any user, tool, or program to find and use content in any
Medicare.gov dataset or GTK content.
 Define what vocabularies are needed to support consumer health care decision making.
 Identify authoritative vocabulary sources for each taxonomy facet.
 Provide vocabularies for each taxonomy facet that are sufficiently defined to be used to build
a functional application (i.e., a CareFinder-like application).
 Be readily extensible to support new application requirements.
 Be flexible enough to accommodate additions of missing categories and changes to existing
categories as needed.
 Define relationships between the vocabularies useful for searching Medicare.gov datasets
and GTK content.

Consumer Health Care Taxonomy: Concept scheme
Eleven facets in the Consumer Health Care Taxonomy displayed in the PoolParty Linked Data
frontend.

Consumer Health Care Taxonomy: Entry terms (skos:prefLabel)
 Entry terms have been identified by analyzing search logs for similar sites, related and
curated content, popular news sources, and user research including uses cases.
 We capture both the technical version of a term and the consumer-friendly or colloquial
version(s) of a term.
 Sometimes the best entry term is the technical version, and sometimes it is the consumer-
friendly version.
Results of Spinal tap v. Lumbar
puncture in Google Fight.

Practical consideration: Post-coordination vs. pre-coordination
 As middleware assisting consumers by reflecting their language, many multiple word concepts
need to be kept together (that is, pre-coordinated) in this taxonomy.
 For example, “Hip fracture” is included in the Taxonomy as a pre-coordinated phrase in the
Conditions facet.

Synonyms and Quasi-synonyms (skos:altLabel)
Variants of “End-stage renal disease”. Quasi-synonyms of “Canes”.

Part of “Hips, legs and feet”. Type of “Fracture”.
Hierarchical relationships (skos:broader, skos:narrower)

Relationships to entry terms in other facets are a custom schema
Semantic Relation Inverse Semantic Relation
Facet Class Facet Class Facet Class Facet Class
Body Locations
and Systems
is affected by Conditions Conditions affects body
location
Body Locations
and Systems
Kidneys is affected by End-stage renal
disease
End-stage renal
disease
affects body location Kidneys
Conditions has treatment of Tests & Treatments Tests & Treatments is treatment for Conditions
End-stage renal
disease
has treatment of Dialysis Dialysis is treatment for End-stage renal
disease
Conditions is concern of Specialty areas Specialty areas is concerned about Conditions
End-stage renal
disease
is concern of Nephrology Nephrology is concerned about End-stage renal
disease
Conditions needs medical
supply
Medical Equipment
& Supplies
Medical Equipment
& Supplies
is needed for
condition
Conditions
End-stage renal
disease
needs medical
supply
Dialysis Equip. &
Supplies
Dialysis Equip. &
Supplies
is needed for
condition
End-stage renal
disease
Care Setting is location for
treatment
Tests & Treatments Tests & Treatments is treatment
provided in
Care Setting
Dialysis Facilities Is location for
treatment
Dialysis Dialysis Is treatment
provided in
Dialysis Facilities

Relationships to entry terms in other facets are a custom schema (2)
Semantic Relation Inverse Semantic Relation
Facet Class Facet Class Facet Class Facet Class
Care Settings specializes in Specialty Areas Specialty Areas is specialty of Care Settings
Dialysis Facilities specializes in Dialysis Services Dialysis Services is specialty of Dialysis Facilities
Medical Supplies &
Equipment
is used in
treatment
Tests & Treatments Tests & Treatments uses medical
supply
Medical Supplies &
Equipment
Dialysis Equipment
& Supplies
is used in treatment Dialysis Dialysis uses medical supply Dialysis Equipment
& Supplies
Specialty Areas includes treatment
of
Tests & Treatments Tests & Treatments is part of practice
area
Specialty Areas
Nephrology includes treatment of Dialysis Dialysis is part of practice
area
Nephrology
Care Settings has focus of
condition
Conditions Conditions is focused on in
setting
Care Settings
Dialysis Facilities has focus of
condition
End-stage renal
disease
End-stage renal
disease
is focused on in
setting
Dialysis Facilities
Body Locations &
Systems
location is treated
by
Tests & Treatments Tests & Treatments treats body
location
Body Locations &
Systems
Kidneys location is treated by Dialysis Dialysis treats body location Kidneys

Semantic relationships diagram
All relationships ESRD relationships

Mapping to Medicare.gov dataset values (skos:relatedMatch)
Example of mapping a Specialty Area to
Medicare.gov data set. (Initial mappings are
narrow in scope).

Project observations
 Consumer healthcare related decision-making behavior is different from clinicians.
 Focus on the problem to be solved: Translate consumer queries into the language necessary for
retrieval of data from Medicare.gov datasets and Good to Know web content.
 Exhaustivity is not a requirement.
 While there are many healthcare-related technical KOS available, consumer-friendly
terminology is generally not available from authoritative sources.
 A lot of work is required to compile a useful KOS from many sources.
 Documentation of editorial guidelines supports this activity and helps to make it scalable.

Project observations (2)
 A small set of extensible taxonomies and custom semantic relationships are sufficient to
develop the domain model.
 A concise set of subject predicate object relationships, e.g., Condition is_concern_of Specialty
Area.
 SKOS is not intended for encoding more complex ontologies beyond thesaurus relationships
(hierarchy, equivalent and generic associative), so a custom schema was developed for specific
associative relationships.
 A strategy to setup separate concept schemes for the Consumer Health Care Taxonomy and
the Medicare.gov datasets controlled vocabularies provided flexibility and extensibility.
 SKOS relatedMatch was used to map across the concept schemes.
 KOS management tools are immature in their capacity to accurately and efficiently batch
import and export KOS, interim taxonomies and semantic relationships.

Resources: Background research
 E. Davenport. “Confessional Methods and Everyday Life Information Seeking.” 44 Annual
Review of Information Science & Technology (2010) pp.533-562.
 Pew Research Center. “Health Fact Sheet.” (December 16, 2013).
http://www.pewinternet.org/fact-sheets/health-fact-sheet/. Last checked October 3, 2016.
 J. Hibbard, S. Sofaer. “Best Practices in Public Reporting No. 1: How to Effectively Present
Health Care Performance Data to Consumers.” AHRQ Publication No. 10-0082-EF. June
2010. http://archive.ahrq.gov/professionals/quality-patient-safety/quality-
resources/tools/pubrptguide1/pubrptguide1.pdf. Last checked October 3, 2016.
 M. Schlesinger, R. Grob, D. Shaller, S. C. Martino, A. M. Parker, M. L. Finucane, J. L. Cerully,
L. Rybowski “Taking Patients’ Narratives about Clinicians from Anecdote to Science.” 373
New England Journal of Medicine (August 13, 2015) pp. 675-679.
http://www.nejm.org/doi/full/10.1056/NEJMsb1502361. Last checked: September 7, 20016.

Resources: Background research (2)
 K. M. Doing-Harris, Q. Zeng-Treitler. “Computer-Assisted Update of a Consumer Health
Vocabulary through Mining of Social Network Data.” 13(2) J Med Internet Res (2011) p. e37.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3221384/. Last checked: September 7, 20016.
http://www.PatientsLikeMe.com/ is a free website where people can share their health data to
track their progress.
 Pew Research Center. “Tracking for Health.” (January 28, 2013).
http://www.pewinternet.org/2013/01/28/tracking-for-health/. Last checked October 3, 2016.

Resources: Center for Medicare and Medicaid Services (CMS) websites
 Hospital Compare. http://www.medicare.gov/hospitalcompare/.
 Nursing Home Compare. http://www.medicare.gov/nursinghomecompare/.
 Physician Compare. http://www.medicare.gov/physiciancompare/.
 Home Health Compare. http://www.medicare.gov/homehealthcompare/.
 Dialysis Facility Compare. http://www.medicare.gov/dialysisfacilitycompare/.
 Supplier Directory. http://www.medicare.gov/supplierdirectory/.
 Data.Medicare.gov. https://data.medicare.gov/.

Resources: Symptom checkers
 Mayo Clinic Symptom Checker. http://www.mayoclinic.org/symptom-checker/select-
symptom/itt-20009075.
 Cleveland Clinic Symptom Checker.
http://my.clevelandclinic.org/health/mysymptomchecker.aspx.
 HealthDirect Symptom Checker. https://www.healthdirect.gov.au/symptom-checker.

Questions?
Joseph Busch (Washington, DC, USA)
jbusch@taxonomystrategies.com
skype: jbusch94110
mobile: 1-415-377-7912

KOS Design for Healthcare Decision-making

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to KOS Design for Healthcare Decision-making

Similar to KOS Design for Healthcare Decision-making (20)

KOS Design for Healthcare Decision-making